Heather Knight

In der aktuellen Ausgabe von Technology Review könnt Ihr ein Interview lesen, das ich mit Robotikerin Heather Knight geführt habe. Ergänzend dazu möchte ich hier im Blog die unbearbeitete englische Original-Version veröffentlichen.

Heather Knight und ihre Roboter. Foto: Pittsburgh Post-Gazette
Heather Knight und ihre Roboter. Foto: Pittsburgh Post-Gazette

Heather Knight promoviert am Robotik-Institut der Carnegie Mellon Universität in Pittsburgh. Ihre Promotion bewegt sich an der Schnittstelle zwischen Robotik und Entertainment. Mit ihrer Firma MarylinMonrobot bringt sie sozial intelligente Roboter auf Bühnen und schafft nach eigenen Worten eine sensorbasierte elektronische Kunst. Heather ist Gründerin des Robot Film Festival sowie des Cyborg Cabaret. 2011 zählte sie das Forbes Magazin zu den 30 wichtigsten Forscher-Persönlichkeiten unter 30.

Why are so many roboticists trying to make machines behave as if they are human? They could do their tasks without being human, maybe even better.

Heather Knight: Robots that operate in isolation have no need to behave like humans or even know we exist. Robots that collaborate with us directly, however, sharing hallways, playing games, helping us clear a table, these robots will need to understand basic social rules and etiquette and how to interpret our behavior.

A unknowingly rude machine might cut someone off going through a door in a crowded hallway. An annoying robot might incessantly repeat its verbal requests when simple gestures would communicate the same more efficiently, e.g., extending its arm during object handoffs instead of saying “I am ready now. Please hand me the object to the right of the red cup.” These situations could all be improved with machines that have a better understanding of human conventions.

So, robot behaviors should become more human-like but not necessarily their physical forms. The form should aid the function — if the robot’s primary occupation involves people, a humanoid form or a capacity for facial expression might make it more relatable. However, if we can achieve modeling of humanlike behaviors without making humanlike forms, I tend to agree with you.

How can robots better understand people, so that humans forget they speak to a machine?

Heather Knight: Researchers have demonstrated that people automatically regard robots as social characters. It makes sense, we see them move autonomously. They have goals. They can change their plans by reacting to objects and people in their environment. It’s how our brains work. It’s theory of mind.

We might think of them as another species, or a range of other species, but in the same way that we’re great at adapting our communications to dogs, cats, adults versus children or even someone that speaks a different language, we will adapt to robots.

But they’ll have to meet us in the middle. Robots should become more human-like in their ability enact natural-seeming behaviors not because they need to, but because we need them to, this is how we interpret each other.

What are the shortcomings of human-machine communications? Are robots better at communicating with people than other kinds of machines?

Heather Knight: Right now, the shortcoming of human-machine communication is that we underestimate our own complexity! Human behavior is what our lives are made up of, so it can be hard to understand why it is so difficult for programmers to turn that into code. It’s like how a foreigner struggles with the grammar of your native language, but you never remember even learning the rules. We have only just scratched the surface of how to make truly effective social machines, in the same way that we are only in the beginning stages of understanding how the brain works.

That being said, the topic is very important, and robots do have certain advantages over other kinds of machines. The physicality of robots makes them uniquely suited for humanlike communication. They have bodies! And share our space! Physically embodied machines captivate us and are more relatable. In person is better than Skype, which is better than email, even in human-human interaction. It’s something about impacting more of our senses – we can touch them, smell them, they are present. This is part of why robots are so popular in Hollywood, they make us care.

Can people bond with machines? Will this lead to a world as we see in „Her“?

Heather Knight: People name their cars. They mourn a lost laptop or corrupted harddrive like a lost loved one. We already bond with machines and that will always happen. But one thing that designing robots makes you intensely aware of is how far away they are from human capability. A baby surpasses the most advanced social robot out there at about 2-months old… it will be a truly long time before machine capabilities compare to humans at the level of friendship. And if they ever do reach that point, we might as well be indistinguishable anyway – from artificial hearts and joint replacements to our co-dependence with our cell phones – society is already full of metaphorical cyborgs.

That being said, I believe it is very important for technology designers and policy makers to be cognizant of the social impact of machines. In research done with elderly people and tele-presence robots, the participants were afraid their children would replace their in-person visits with visits through robots, they thought that would be awful. The idea that they could use the robots to “visit” their kids and see their grandchildren using the robots themselves, however, was enthusiastically welcomed. They even thought it would be nice to use the robots “attend” an outdoor concert “in person,” as they missed being able to do that by themselves. The goal should be to make technology that makes humanity flourish, enabling their personal freedoms.

What is the key to making human-like machines? voice, mimicry, movement?

Heather Knight: Lately, I’ve been working on robot movements. There’s a saying that 95% of communication is body language, and I do think that non-verbal channels of communication can make more subtle, natural and effective interactions. An old colleague of mine Guy Hoffman gave a talk ending with robots that play music. He found that giving one of the robots a bobbing and turning head during collaborative performances helped its partner to understand its intentions (e.g. when to end the piece), but most of all, people liked that it appeared to be having a good time.

In the end, all the channels of communication can come together, but depending on the application, we can pick and choose. Maybe a robot delivering samples in a hospital can be more like a dog – diligently working, friendly and enjoying praise when its has done its job well. That sort of robot probably does not require natural speech, and could be more like the furniture that comes alive in Beauty and the Beast.

Can using robots as actors somehow help to create robots who behave and look more human?

Heather Knight: Guy Hoffman was one of the people that made me start thinking in that direction, back when we were sharing an office at the MIT Media Lab in 2007. Coming from a programming background, he was using acting theory to think about how to make more intuitive and impulsive AI systems. I, on the other hand, started out as a mechanical engineer, so I’ve become fascinated with adapting some of the physical expressiveness of the stage to generating natural and charismatic machines.

Performers are the best-trained artificial emotion systems out there. Actors manufacture the experience of authentic characters that manage to capture our imaginations, and I mean that as a complement. They absolutely have a great deal to offer in the creation of, as you describe it, more humanlike machines. The challenge is, of course, to make their body of knowledge algorithmic. Stay tuned!

Foto: Pittsburgh Post-Gazette